AI can reveal new cell biology just by looking at pictures

Artificial intelligence has learned how to recognize and categorize different dog breeds from photos. A new machine learning method from CZ Biohub now enables classification and comparison of different human proteins from fluorescent microscopy images. Credit: CZ Biohub

Humans are good at looking at pictures, finding patterns, or making comparisons. Look at a bunch of dog pictures, for example, and you can sort them by color, ear size, face shape, etc. But can you compare it quantitatively? And perhaps most interestingly, can a machine extract meaningful information from images that humans cannot?

Now, a team of Chan Zuckerberg Biohope scientists from Stanford University has developed a machine-learning method to quantitatively analyze and compare images – in this case microscopic images of proteins – without prior knowledge. as stated in Nature’s WaysTheir algorithm, dubbed “cytoself,” provides rich and detailed information about the protein’s location and function within the cell. This ability could speed up research time for cell biologists and eventually be used to speed up drug discovery and drug screening.

“This is very exciting – we’re applying AI to a new kind of problem and we’re still recovering everything humans know, plus more,” said Loic Royer, co-author of the study. “In the future we can do this for different types of images. It opens up a lot of possibilities.”

Cytoself not only demonstrates the power of machine learning algorithms, but has also generated insights into cells, the building blocks of life, and, in proteins, the molecular building blocks of life. Each cell contains about 10,000 different types of proteins – some that work alone, many work together, and do different jobs in different parts of the cell to keep them healthy. “The cell is more spatially organized than we thought before. This is an important biological finding about how the human cell is wired,” said Manuel Leonetti, also a co-author of the study.

And like all the tools developed at CZ Biohub, cytoself is open source and available to everyone. “We hope it will inspire a lot of people to use similar algorithms to solve their own image analysis problems,” Leonetti said.

No matter a PhD, machines can learn on their own

Cytoself is an example of what is known as self-learning, meaning that humans do not teach the algorithm anything about protein images, as in supervised learning. “In supervised learning, you have to teach the machine one by one with examples; it’s a lot of work and it’s very tedious,” said Hirofumi Kobayashi, the study’s lead author. And if a machine is limited to classes that humans teach, it can introduce bias into the system.

“Manu [Leonetti] “I think the information was already in the photos,” Kobayashi said. And we wanted to see what the machine could detect on its own.

In fact, the team, which also included CZ Biohub software engineer Keith Cheverals, was surprised by how well the algorithm was able to extract it from the images.

“The degree of detail in the protein’s localization was much higher than we thought,” said Leonetti, whose team has developed tools and techniques for understanding cell structure. “The machine converts each protein image into a mathematical vector. And then you can start arranging the images that look similar. And we realized that by doing that, we could predict, with high accuracy, which proteins are working together in the cell just by comparing their images, which was surprising Somewhat “.







In this rotating 3D UMAP image, each dot represents a single protein image, colored according to protein localization categories. Collectively it constitutes a highly detailed map of the entire diversity of protein localization. Credit: CZ Biohub

first of its kind

Although there has been some previous work on protein images using self-supervised or unsupervised models, self-supervised learning has never before been successfully used on this large data set of more than 1 million images covering more than 1,300 proteins measured from cells live human. Kobayashi is an expert in machine learning and high-speed imaging.

The images were a product of CZ Biohub’s OpenCell, a project led by Leonetti to create a complete map of the human cell, eventually including the characterization of the 20,000 or so types of proteins that power our cells. Published earlier this year in Sciences They were the first 1,310 proteins that were tagged, including images of each protein (produced using some type of fluorescent tag) and mappings of their interactions with each other.

Cytoself was key to the achievement of OpenCell (all images available at opencell.czbiohub.org), providing very accurate and quantitative information about the localization of the protein.

“The question of all the possible ways a protein could be located in the cell – all the places it could be and all kinds of combinations of places – is fundamental,” Royer said. “Biologists have tried to identify all the possible places, for decades, all the possible structures within the cell. But this has always been done by humans looking at the data. The question is, how much do human limitations and biases make this process imperfect?”

Royer added, “As we’ve shown, machines can do this better than humans can do. They can find finer categories and see differences in images that are pretty cool.”

The team’s next goal for cytoself is to track how small changes in protein localization can be used to learn about different cellular states, for example, a normal cell versus a cancer cell. This may be the key to better understanding many diseases and facilitating drug discovery.

“Drug screening is basically trial and error,” Kobayashi said. “But with cytoself, this is a huge leap because you won’t need to do experiments one by one with thousands of proteins. It’s a low-cost method that can speed up research a lot.”


AI software accurately predicts protein localization


more information:
Hirofumi Kobayashi et al, Self-supervised deep learning encodes high-resolution features of subcellular protein localization, Nature’s Ways (2022). DOI: 10.1038 / s41592-022-01541-z

Presented by Stanford University

the quote: AI can reveal new cell biology just by looking at images (2022, August 1) retrieved August 1, 2022 from https://phys.org/news/2022-08-ai-reveal-cell-biology-images .html

This document is subject to copyright. Notwithstanding any fair dealing for the purpose of private study or research, no part may be reproduced without written permission. The content is provided for informational purposes only.


#reveal #cell #biology #pictures

Leave a Comment

Your email address will not be published.